The Neuroanatomy of Speech Sequencing at the Syllable Level Feng Rong1, A. Lisette Isenberg1, Erica Sun1, Gregory Hickok1, CA 1. Department of Cognitive Sciences, University of California, Irvine. Irvine, CA 92697. [email protected] Grant support: NIDCD R01 DC009659 ABSTRACT Correctly ordering a sequence of speech sounds is a crucial aspect of speech production. Although studies have yielded a rich body of data on the neural substrates of visuomotor sequencing and sequence learning, research on brain regions and their functions involving speech sequence production hasn’t attracted much attention until recently. Previous functional MRI studies manipulating the complexity of sequences at the phonemic, syllabic, and suprasyllabic levels have revealed a network of motor-related cortical and sub-cortical speech regions. In this study, we directly compared human brain activity measured with functional MRI during processing of a sequence of syllables compared with the same syllables processed individually. Among a network of regions independently identified as being part of the sensorimotor circuits for speech production, only the left posterior inferior frontal gyrus (pars opercularis, lIFG), the supplementary motor area (SMA), and the left inferior parietal lobe (lIPL) responded more during the production of syllable sequences compared to producing the same syllables articulated one at a time. Introduction Fluent speech requires the rapid coordination of vocal tract gestures to produce the intended sequence of phonemes (segments), syllables, and words. Speech error data (slips of the tongue, Spoonerisms) have illuminated the process, providing direct evidence for sequence planning over multiple representational levels (Dell, 1995; Fromkin, 1971; Garrett, 1975; W. J. M. Levelt, 1989; Shattuck-Hufnagel, 1992). The following examples illustrate sequence errors at the phoneme, syllable, and word levels (Fromkin, 1971) (intended utterance actual (slip) utterance): phoneme: keep a tapeteep a kape syllable: philosophyphi-so-lo-phy; butterfly and caterpillarbutterpillar and caterfly word: a computer in our own laboratorya laboratory in our own computer Such evidence from natural slips of the tongue, laboratory induced slips (Baars, Motley, & MacKay, 1975), chronometric studies of object naming (Schriefers, Meyer, & Levelt, 1990), computational modeling (Dell, 1986), and speech error data following brain injury (Dell, Schwartz, Martin, Saffran, & Gagnon, 1997; Walker & Hickok, 2016) has led to much progress in understanding the cognitive mechanisms behind the sequencing of speech sounds during language production (Dell, 1986; Dell, 1995; Dell et al., 1997; W. J. M. Levelt, 1989, 1999). A number of studies have investigated the neural foundation of speech production using a range of methods with notable progress in mapping the broad stages of speech (e.g., lexical-semantic versus phonological) onto neural networks (Dell, Schwartz, Nozari, Faseyitan, & Branch Coslett, 2013; Indefrey, 2011; Indefrey & Levelt, 2004; Peeva et al., 2010; Price, 2012; Wilson, Isenberg, & Hickok, 2009). Another line of research has made significant progress in understanding the role of sensorimotor circuits in speech motor control (Golfinopoulos, Tourville, & Guenther, 2010; Guenther, 2006, 2016; Hickok, 2012; Hickok, Houde, & Rong, 2011; Houde & Nagarajan, 2011; Tian & Poeppel, 2015; Walker & Hickok, 2016). Overall these studies have identified a distributed speech production network that includes pre- and post-central gyri, medial premotor cortex (SMA/pre-SMA), lateral premotor cortex, posterior inferior frontal gyrus, anterior insula, superior temporal gyrus, and the posterior planum temporale region, termed Spt (Hickok, Buchsbaum, Humphries, & Muftuler, 2003; Hickok, Okada, & Serences, 2009), as well as portions of the cerebellum and basal ganglia (Guenther, 2016). Relatively few studies, however, have explicitly studied the neural circuits that support speech sequencing, an endeavor that could eventually link psycholinguistic and neural models of speech planning. One functional MRI study that did so (Bohland & Guenther, 2006) manipulated sequence complexity (number of unique syllables in the set: ta-ta-ta vs. ka-ru-ti) and reported activations associated with greater complexity in a network including pre-SMA, frontal operculum/anterior insula (bilaterally), lateral premotor cortex and the posterior inferior frontal gyrus/sulcus (left lateralized). One limitation of this study, however, is that it is impossible to know whether the activations are driven by the sequencing demands per se or simply by the increased complexity demands associated with articulating different tokens, independently of whether the sequence is correctly produced. For example, repeating the same token three times versus three different tokens should lead to a difference in the degree of neural adaptation in regions coding motor plans for syllables independently of the sequencing demands. This study, along with others (Shuster & Lemieux, 2005), also manipulated syllable complexity (number of segments within the syllable), which should increase segment sequencing load; a similar network was implicated. But again, because syllable complexity manipulations involve a different number of phonemes produced, it cannot definitively isolate sequencing per se as opposed to simply more time spent planning articulation. One fMRI study (Peeva et al., 2010) moved toward avoiding these confounds by contrasting several articulatory conditions, one in which the same item was produced repeatedly (PIGRA PIGRA PIGRA PIGRA...), a second with two alternating items using the same phonemes but with reordered syllables (ZE.KLO KLO.ZE ZE.KLO KLO.ZE...), a third with two alternating items using the same phonemes but in two different syllabification patterns (LO.FUB FU.BLO LO.FUB FU.BLO…), and a fourth with variable phonemes and syllables (GUPRI DRAVO VIBAG NUVAF…). Activation patterns to the various conditions differentiated some of the previously identified regions involved in syllable complexity/sequencing. Specifically, the left SMA responded similarly to the first three conditions and greater to the fourth (variable) condition; the ventral premotor cortex responded similarly to the first two conditions, greater to the third (alternating syllabification pattern) condition, and most to the fourth (variable) condition; and the right cerebellum responded least to the first condition, more to the second (alternating syllable order) and third (alternating syllabification pattern) conditions, which did not differ from one another, and most of the fourth (variable) condition. The authors concluded that (i) the left SMA codes speech information at the phoneme level, thus maximally activating only when there is variation in the articulated phonemes, (ii) the left ventral premotor cortex codes speech information at the syllable level, thus modulating activity as a function of variation in the syllable structure, and (iii) the right cerebellum is coding speech information at the suprasyllabic level, thus modulating activity as a function of variation in syllable order (or structure). This finding and interpretation challenges the idea that the ventral premotor cortex and SMA play a particular role in syllable sequencing. However, as this study involved overt speaking, which is known to modulate activity in motor related areas (Hickok et al., 2003; Hickok et al., 2011; Wilson, Saygin, Sereno, & Iacoboni, 2004), it is unclear whether the different activity profiles are influenced more by response properties during speech planning or a mixture of planning and perception, which could vary from one region to the next (although (Bohland & Guenther, 2006) also used overt speech). Yet another approach to mapping the network involved in speech sequence processing is to identify regions involved in learning novel sequences. Segawa and colleagues (Segawa, Tourville, Beal, & Guenther, 2015) compared fMRI activation patterns to novel versus previously learned non-native (phonotactically illegal) phoneme sequences and reported greater activation for novel sequences in premotor cortex, including both dorsal and ventral (par opercularis) clusters, the frontal operculum (FO), the superior parietal lobule, posterior superior temporal gyrus (pSTG) and posterior superior temporal sulcus (pSTS), inferior temporal-occipital cortex, and globus pallidus. One complication in interpreting these results is that the duration spent articulating learned and novel sequences differed, making it hard to attribute to activation differences to the sequence processing. An additional analysis showed that the frontal operculum activity was significantly correlated with learning success, which alleviates this concern, and suggests that the FO may play an important role in speech sequencing. The present study sought to map the network involved in speech sequence planning without the potential confounds of overt sensory responses, while taking a different approach to avoiding confounds associated with differing numbers of syllable types or phonemes in the contrasting conditions. Our approach involved the auditory presentation of four different syllables for both the sequence and the non-sequence conditions. In the sequence conditions, the four syllables were presented immediately one after the other, after which the participant covertly repeated the entire sequence. In the non-sequence (“unit”) condition, a short silent interval occurred after each